[Bug tree-optimization/18487] Warnings for pure and const functions that are not actually pure or const

[dmalcolm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=18487

David Malcolm <dmalcolm at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |UNCONFIRMED
                 CC|                            |dmalcolm at gcc dot gnu.org
         Resolution|WONTFIX                     |---
     Ever confirmed|1                           |0

--- Comment #9 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #8)
> I don't think I agree with closing this as won't fix as shown now we have
> three duplicated bugs asking the same thing.

I agree; reopening.

It may not be possible to implement a perfect test for this, but it seems to me
that it's possible to implement a warning that catches a subset of cases
without false positives (at the cost of having false negatives).

As a first iteration of an implementation:

  for all fns labelled as pure/const:
    for all basic blocks that are guaranteed to be in a
         path through the function:
      for all stmts in BB:
        if stmt is non-pure/non-const:
          issue a warning

(yes, this would miss some things e.g.
   if (COND)
     non-pure
   else
     non-pure
but could be cheap and effective)